By Andrew Goodman, 3/15/2003

Last week, with no small fanfare, Google announced a new distribution model for their AdWords keyword-based text advertising: content targeting. For existing Google advertisers this was an unexpected development. Although advertisers' ads will be triggered by the keywords they're bidding on, content-targeted ads appear in a venue which is significantly different than search results: namely, next to articles, weblog entries, or newsgroup postings. Advertisers are allowed to opt out of Content Targeting if they wish.

The content targeting project was quietly piloted in the fourth quarter of 2002 on small sites such as Howstuffworks.com and Weather Underground, as well as on Google Groups. Google stresses that they're using more advanced targeting technology than some of their competitors have used to place ads next to content. A common method, of which Google is critical, is to simply categorize sites into broad "channels," and serve up ads from advertisers who are bidding on generic keywords relating to those channels.

Overture, it appears, has been experimenting with some contextually-based advertising of its own. Overture ads for Boston-area hotels appear on a Yahoo! Weather page for Boston, for example. In other cases, it appears that Overture's attempts to place ads next to content are at the experimental stage. Minneapolis-based consultant Ed Kohler, whose company is called Haystack in a Needle, saw a clickthrough in his log file which had been triggered by his Overture ad which turned out to have appeared near a completely unrelated music search on Yahoo's Launch.com service.

While it's not clear how Overture is attempting to match content with ads, it's not much clearer exactly how Google is doing it, either. Susan Wojcicki, a product development manager for Google Content Targeting, points out that Google's choice of relevant advertising is done dynamically, "on-the-fly," by analyzing the entire content on a page and matching bidders' keyword-triggered ads with that content. Google, presumably, has an advantage here because of the large number of content pages that already exist in its search index. Unlike many early-generation targeting methods, matches here may be highly granular. Instead of just matching advertisers for "shoes" with pages about "shoes," Google's technology also aims to match advertisers for highly specific items like "vintage bowling shoes" with content about just that. All that being said, there is still a lot that is mysterious about the process.

Wojcicki argues that the ongoing shift from intrusive graphical skyscraper ads to micro-targeted keyword-based ad units from Google "will improve the overall user experience on the web" due to the "extreme relevance" of the advertising. One might also say that this represents a new opportunity for publishers who have been having trouble monetizing content and managing ad sales. However, this remains to be seen. Google's business model at this stage doesn't involve a revenue share with publishers. Rather, Google makes a CPM-based offer for a large "media buy" of ad space, and pays the publisher that rate while collecting revenue on the clicks. This might be skewed against publishers insofar as their upside is limited while Google benefits from rising per-click costs for this form of advertising. Google takes a risk, too, though. They stand to lose money if clickthrough rates are abysmally low and they can't recoup the initial ad buy.

And the fact is, even with the hyper-targeting, clickthrough rates might be very low. Reading a novel excerpt online describing someone's "slightly out-of-date Calvin Klein dress shirt" is not nearly as action-oriented as typing "Calvin Klein dress shirt" into a search engine. In the latter case, the user entered those words - they are literally the user's creation. In the former case, the author and publisher put those words on the page, and the user, while he may be vaguely interested, he is far less likely to interrupt what he's doing (reading fiction) to click on an ad. That premise is being borne out in the early going as the clickthrough rates on content-targeted ads look to be significantly lower than CTR's for ads appearing next to Google search results.

Granted, not everything you see online is a novel excerpt to be passively read by a "surfer." Specialized trade publications and highly granular subjects like weather are likely to offer a better backdrop for relevant commercial messages. No doubt this is why About.com's Sprinks advertising service launched ContentSprinks, keyword-based ads appearing in the online versions of Primedia trade magazines, and why rivals like Overture, Findwhat, Search123, and Revenue Software have all been exploring the content targeting model.

While clickthrough rates might indeed be lower, Google claims that their tests show that post-click behavior (conversions to sales) resulting from content-targeted ads is similar to that seen with search engine advertising. Thus, no one in particular is harmed by the low CTR's assuming there are a large number of page impressions served daily and assuming the ads don't annoy users too much.

Industry reaction to Google's announcement has been lukewarm. Experienced reporters are asking, rightly, "hasn't content targeting been the whole goal of online advertising for several years?" It seems clear that Google's entry into this market is an evolutionary step, not a revolutionary advance - although it has sobering implications for traditional ad middlemen like Doubleclick who have already ceded a sizeable chunk of the overall web advertising pie to Google and Overture.

According to Gil Elbaz, co-founder and CIO of Applied Semantics, a meaning-based search technology firm which offers ad targeting technologies called AdSense and DomainSense, "for one reason or another, past ad targeting efforts have been flawed."

Applied Semantics, which drives traffic to advertisers' sites through partnerships with Overture, Findwhat, and individual publishers, believes it brings better ad targeting to the table than Google offers. For the time being, Applied Semantics' offerings differ from Google's in several key ways, some of which might prove important to publishers. One difference that doesn't relate to the technological side of the equation is that Applied Semantics pays publishers on a revenue-share basis rather than negotiating CPM-based media buys as Google says it will be doing.

Part of the difference is that Applied Semantics has focused its entire business on developing a proprietary categorization database that understands the relationships between words and concepts. Along with lesser-known providers of semantic technology to the enterprise (such as H5 Technologies, which began its life under a development code name, ejemoni), Applied Semantics can read and "understand" the meaning of concepts on a page. Many lesser matching technologies are likely using rudimentary keyword matching. Applied Semantics' database, which is updated under the supervision of lexicographers, contains 1.25 million terms with "tens of millions of relationships amongst them," says Elbaz.

Elbaz also offered some conjecture about how Google's technology works "based on some industry talk and our guess as to what they're up to." Essentially, along with some keyword matching, "they're probably using some kind of user tracking, looking at statistics about what readers on content pages tend to click on."

Applied Semantics hopes that the recent interest in content targeting will create more interest in "categorization and semantic analysis" as another means of improving the relevancy of ranked search results. The very fact that search engine algorithms remain largely keyword-based means that they aren't particularly sophisticated in learning what a page is "about." According to Elbaz, semantic researchers such as these Stanford University authors have argued that current search algorithms are rapidly approaching a "ceiling" of relevancy. But there is talk in semantic research circles of a "new higher ceiling" which would be made possible, for example, by the use of semantic analysis to classify search engine spam.

Users, of course, hate spam wherever they see it. Irrelevant search results, unsolicited emails, and poorly-targeted, intrusive ads are all turnoffs, notwithstanding the protestations of some online ad agency dinosaurs who still believe that intrusive equals effective.

Although targeting technology is far from new, Google's announcement has placed it in the forefront again. And whether such targeting is ultimately provided by a search giant like Google, a multi-channel online ad middleman firm like Doubleclick, or a laser-focused upstart like Applied Semantics, it's clear that the push towards more intelligent targeting is going to improve the user experience and, publishers hope, shore up ad revenues enough to make free online content a worthwhile business model.

From the pay-per-click keyword advertiser's standpoint, content targeting represents an interesting development. But for now, most are concluding that when it comes to finding interested consumers, nothing beats advertising next to search results.